De Nederlandsche Bank and AFM expectations and AI operations careers in the Netherlands

Two supervisors shape AI operations work in Dutch financial firms. De Nederlandsche Bank looks at soundness, risk management and integrity; the AFM looks at how customers are treated and how markets behave. An agent in a Dutch bank, insurer or pension provider must satisfy both, so firms need people who can review its decisions against each lens and record why.

What do DNB and the AFM expect of people who oversee agents?

Neither supervisor has written a separate AI rulebook for front-line staff. Both apply existing duties to AI: sound governance, controlled operations, careful treatment of customers, and responsibility that stays with the firm and its management. Both have published views on responsible AI use in finance that stress, in substance, accountability, explainability and human involvement where outcomes matter.

For you, the two lenses translate into two questions on every review. The DNB question: is this action within the agent's controlled bounds, and would the firm's risk function recognise the decision? The AFM question: is this outcome fair and clear for the customer, and would the customer understand it?

The EU layer sits underneath. DORA, applying from 17 January 2025, covers ICT risk, incidents and third-party providers. The EU AI Act asks deployers of high-risk systems, such as credit scoring of individuals or pricing for life and health insurance, to give human oversight to people with the competence, training and authority to do it. GDPR covers the data, supervised in the Netherlands by the national data protection authority.

Who is covered?

DNB supervises banks, insurers, pension funds and payment institutions on the prudential side, and checks integrity matters such as anti-money laundering controls at many of them. The AFM supervises conduct across financial services, from lenders and insurers to investment firms and advisers. The largest banks also fall under ECB banking supervision.

Dutch pension funds deserve a note. They run large administrative operations where agents can help with member correspondence and data checks, and they sit under DNB's supervision. That makes them a less obvious but real employer of oversight staff.

Firms outside finance answer to the EU AI Act and GDPR but not to DNB or the AFM.

Which roles and competence does this create?

Dutch adverts may be in Dutch or English, and titles vary: medewerker klantcontact, operationeel analist, kwaliteitsmedewerker, KYC analyst, first-line risk officer. Read the duties for review, exceptions, approvals, quality checks and escalation rather than looking for one title.

The roles that tend to carry oversight duties are:

• Customer servicing reviewer. Approves agent-drafted replies and account changes against AFM expectations of clarity and fairness. Many come from branch and contact centre work; see AI operations careers for bank tellers in the Netherlands.
• KYC and integrity reviewer. Checks agent-prepared customer due diligence files and alerts, where DNB's integrity supervision applies. The path from compliance roles is in AI operations careers for compliance analysts in the Netherlands.
• Credit or claims reviewer. Approves agent-held decisions in lending and insurance, the areas most likely to be high-risk under the EU AI Act.
• First-line risk and quality analyst. Samples decisions, tests controls and reports to the agent owner.
• Agent owner. Accountable for one agent's scope, sources and certification.

The skill employers look for is the ability to hold both lenses at once, and to write a decision note that a risk officer and a conduct supervisor would each accept.

How can you show employers you meet both lenses?

The competence side is explored further in De Nederlandsche Bank and AFM expectations for AI agent oversight skills.

What might a Dutch interviewer ask?

Dutch interviews are often direct and value plain answers. Expect questions such as:

1. "The agent approves a loan the customer can technically afford but which leaves them almost nothing each month. Is that fine?" Show you see the conduct question, not only the rule check, and that you would hold or escalate it.
2. "How would you explain the agent's decision to the customer?" Give the reasons in ordinary Dutch or English, without system jargon.
3. "What would you record if you overrode the agent?" Name the facts you relied on, the bound or policy involved, and your name.
4. "How would you feel about your decisions being logged?" Many Dutch firms have a works council that discusses such logs. Say that the log exists so decisions can be reconstructed, and that you would expect its use to be agreed.

A candidate we will call Sanne, a pension administration officer, prepared by writing out how she would review an agent's draft letter to a member about a transfer. It let her answer both the soundness and the fairness question with one real example.

What do these expectations not cover?

They do not certify individuals and do not require a specific course. They do not reach non-financial employers. They do not settle employment-law questions about how decision logs are used, which depend on works council agreements and contracts.

They also do not mean the Dutch approach matches every other EU market. Compare the Irish picture in Central Bank of Ireland expectations and AI operations careers in Ireland, where the EU rules are shared but supervision is organised differently.

This is an explanation of the rules, not legal advice; confirm how they apply to you with counsel.

Where the Academy fits

Pryme Intelligence Academy lets you build evidence before you have the job. Foundations is free and covers why agents must be governed. Certified Pryme Operator adds a written exam and a two-week practicum; without an employer, you do it on a simulated queue in a free workspace. The certificate is issued through VerifyDoc.ai and listed on a public register with its expiry.